Pakistan-Linked APT36 Deploys AI-Generated 'Vibeware' Against Indian Government in First Documented Nation-State Use of Vibe-Coded Malware
Bitdefender documents APT36 using LLMs to mass-produce malware in Nim, Zig, and Crystal at a daily cadence, flooding Indian government networks with disposable implants in a strategy researchers call 'Distributed Denial of Detection.'
Overview
Bitdefender has documented what it describes as the first confirmed instance of a nation-state threat actor systematically using AI-generated code to produce malware at industrial scale. The Pakistan-linked group APT36, also known as Transparent Tribe, has been deploying waves of AI-written malware — dubbed “vibeware” — against Indian government networks, diplomatic missions, and entities in South Asia, according to Hackread, which reported the findings exclusively on March 5.
The campaign’s distinguishing feature is not sophistication but volume. Rather than crafting targeted, high-quality implants, APT36 is using large language models to rapidly generate functional malware in niche programming languages including Nim, Zig, Crystal, Rust, and Go — languages that many traditional antivirus engines are poorly equipped to analyze. Bitdefender estimates the group is producing new variants at a daily cadence, a strategy researchers have labeled “Distributed Denial of Detection,” according to CyberInsider.
How Vibeware Works
The term “vibeware” derives from “vibe coding,” the practice of describing desired functionality to an AI model and having it generate the code. In this context, APT36 appears to be using LLMs to port existing malware logic into unfamiliar languages with minimal human effort. The AI collapses the expertise gap — an operator who has never written a line of Nim or Crystal can produce functional malware by describing the desired behavior in natural language, according to SecurityBrief.
The resulting code is not technically impressive. Bitdefender found samples with basic errors: one credential-stealing tool contained a placeholder instead of a command-and-control server address, meaning it could never actually exfiltrate data. Another backdoor’s status-reporting function reset its own timestamp on every call, causing the infected host to always appear online regardless of its actual state, according to Hackread.
But quality is not the point. By flooding target environments with dozens of disposable variants written in languages that signature-based detection engines rarely encounter, APT36 overwhelms defenders through sheer volume. Even if 90 percent of the samples are caught, the remaining 10 percent achieve persistence — and the cost of generating replacements is trivially low.
The Malware Arsenal
Bitdefender identified several specific tools in the campaign. CrystalShell, written in Crystal, uses Discord channels as a command-and-control interface. SheetCreep converts Google Sheets into a C2 hub for issuing instructions and receiving exfiltrated data. LuminousCookies is designed to bypass App-Bound Encryption in Chrome and Edge browsers by injecting into browser memory. BackupSpy scans connected drives and USB devices for specific file types, according to CyberInsider and Hackread.
Across the campaign, the group uses “Living Off Trusted Services” (LOTS) techniques, routing communications through Discord, Slack, Google Sheets, Firebase, and Supabase — legitimate cloud platforms that are difficult to block without disrupting normal business operations, per CyberInsider.
Targets and Attribution
The primary targets are Indian government agencies and diplomatic missions across multiple countries, with secondary targeting of entities in Afghanistan and private-sector organizations. Attackers appear to be seeking sensitive information related to military personnel, diplomatic communications, and national security policies, according to CyberInsider.
Bitdefender attributes the campaign to APT36 with medium confidence, based on infrastructure overlaps and tool continuity with previously documented Transparent Tribe operations, per GovInfoSecurity.
The Broader Implications
The vibeware campaign represents a qualitative shift in how AI affects the threat landscape. Rather than enabling fundamentally new attack techniques, LLMs are industrializing the production of existing ones. The barrier to generating malware in any given programming language has collapsed, making language diversity a viable evasion strategy for the first time.
This creates an asymmetric burden. Defenders must maintain detection capabilities across an expanding range of languages and frameworks, while attackers can generate new variants in minutes. Traditional signature-based antivirus — still the primary defense layer for many organizations — is structurally disadvantaged against an opponent that can produce unique binaries at negligible cost.
What We Don’t Know
Bitdefender has not disclosed the total number of unique vibeware samples identified, the specific LLM or LLMs used by APT36, or whether the campaign has achieved any confirmed data exfiltration from its targets. The medium-confidence attribution leaves open the possibility that a different group with overlapping infrastructure is responsible. Whether other nation-state actors are adopting similar AI-assisted malware production pipelines has not been publicly documented, though the technique’s low cost and high effectiveness make broader adoption likely.